In　this　paper,　we　study　the　Bregman　kappa-means　problem　with　respect　to　mu-similar　Bregman　divergences　(p-BKMP).　Given　an　n-point　set　S　and　kappa　＜=　n,　mu-BKMP　is　to　find　a　center　subset　C　subset　of　S　with　vertical　bar　C　vertical　bar　=　k　and　separate　the　given　set　into　k　clusters　accordingly,　aiming　to　minimize　the　sum　of　mu-similar　Bregman　divergences　of　the　points　in　S　to　their　nearest　centers.　We　propose　a　new　variant　of　k-means++　by　employing　the　local　search　scheme,　and　show　the　algorithm　deserves　a　constant　approximation　guarantee.